When the Numbers Aren't In Yet: Smart Shortcuts for Early Project Decisions

Around the world, large capital projects force owners to make the decisions that matter most while the information that would make those decisions comfortable is still missing. Concept choice, early cost and schedule ranges, partner selection and the first public narrative all arrive before detailed engineering has produced reliable numbers. Some teams respond by inventing precision. Others freeze and wait for data that will come too late to change the concept. Neither response serves the project. This comprehensive blog will explore how owner teams on large capital projects can decide well when the numbers are not in yet, how to tell a useful heuristic from a biased shortcut, why simple early methods often beat fake detail, which small kit of rules belongs at the concept gate, and when a shortcut should be handed off to another tool entirely.

1. When the Board Asks for a Number That Does Not Exist Yet

Consider a familiar scene. An investment committee sits down to review a major scheme. The preferred concept is sketched. The site is known in outline. Analogous projects elsewhere are remembered, vaguely. Detailed design is months away, and the front-end engineering that would produce a serious estimate has not begun. Yet the pack in front of the committee still contains a single cost, a single date and a single return, each carried to a reassuring number of decimal places.

Two failure modes follow from that demand. In the first, the owner team manufactures precision. Spreadsheet rows multiply. Contingencies are buried rather than shown. The story becomes clean, and subjective confidence rises with the cleanliness of the story rather than with the quality of the evidence. In the second, the team refuses to decide at all. "We need better data" becomes a reason to skip appraisal, to leave the concept unchallenged, or to let a preferred scheme lock by default while everyone waits for numbers that cannot yet exist.

Both patterns misunderstand the front end of a large capital project. The decisions that lock the most long-term value — which concept to pursue, whether the scheme serves strategy, which fatal downsides rule an option out — are made while information is still scant. That is not an aberration. It is the normal condition of early work. Planning effort that arrives only after go-ahead mainly improves the tactical delivery of whatever concept was already chosen. The strategic question has to be answered earlier, in thinner air.

The pressure that produces these patterns is real. Sponsors need something that can travel through a board pack. Financiers ask for a number they can model. Public narratives prefer a date. None of those audiences is wrong to want clarity. The error is to confuse the clarity of a single figure with the clarity of a decision. A range that names its drivers is often clearer than a point estimate that hides them. A concept choice that states what would kill it is clearer than a preferred scheme defended by decorative decimals.

The central claim of this blog follows from that condition. On large capital projects, waiting for perfect numbers is how the wrong concept locks. Inventing perfect numbers is how optimism dresses up as analysis. What owner teams need instead is a small kit of adaptive heuristics: fast, frugal rules that use a few decisive cues well, stop searching when those cues have spoken, and stay honest about what is still unknown. Those rules are not a licence for sloppiness. They are the disciplined alternative to both paralysis and fake precision.

2. Scant Information Is Not a Defect

Early-phase work on large capital projects runs on qualitative cues, analogies, hearings, lessons from comparable schemes and rough ranges. Creativity, structured judgement and challenge belong at this stage. The absence of rich quantitative data is not an excuse for skipping appraisal, and it is not a licence for pretending the data exist.

A useful way to see the trap is to notice where planning money usually goes. Most of it is spent on detailed planning after a go-ahead decision. That spend improves the chance of delivering the chosen outputs on time and budget — tactical performance. Investing more before go-ahead aims at a different target: strategic performance, meaning a concept that is relevant, effective and worth the life-cycle cost. The quality of that early identification, preparation and appraisal process is what decides whether the project is the right one, not merely whether it will be built efficiently.

Uncertainty is highest when a concept is conceived, and it falls as information accumulates. Decision-maker flexibility mirrors that pattern: it is highest early and collapses as choices lock. The practical implication is that concept choice should be forced while amendment is still cheap, using the information that is actually available — including scant and primarily qualitative information — rather than deferred until detail arrives and flexibility has already been spent.

There is a complementary warning about premature quantification. At concept stage, more spreadsheet rows do not equal more truth. Early cost and schedule forecasts that lean on parametric methods and class analogies are often the honest form of early numbers. Forcing FEED-grade detail before the concept is settled creates the appearance of diligence and the substance of error. Scant information, used deliberately, is a filter. Fake detail is a fog.

This is also why "we lack data" is a dangerous sentence at the front end. Lack of detailed data is expected. What is not acceptable is using that lack as a reason to skip the strategic work that scant-information methods are designed for: challenging the concept, testing needs against strategy, comparing classes of similar schemes, and keeping options open while flexibility still exists. The early phase is exactly when qualitative hearings, lessons, expert challenge and rough stochastic ranges earn their keep. Treating them as inferior to a spreadsheet that invents quantities is a category mistake.

3. Bias Versus Useful Heuristic

Fast shortcuts are not all the same, and owner teams get into trouble when they treat them as if they were.

Some shortcuts are the biased habits of the fast mind working under incomplete information. The mind builds the most coherent story it can from whatever is currently visible and does not naturally ask what evidence is missing. Confidence then tracks the quality of the story, not the quality of the evidence. Anchoring on the first number seen, overweighting what is easy to recall, treating the project as unique and therefore incomparable, and expressing more certainty than the evidence supports all flow from that habit. Those patterns have been mapped elsewhere as a system problem — individual, group and organisational — and they remain a constant risk at every gate.

Other shortcuts are adaptive. Under complexity, limited time and limited information, simple heuristics that ignore most of the available noise and use a few cues well can outperform optimisation routines that pretend the model is complete. Recognition heuristics ask whether a class of similar schemes is known at all. Lexicographic rules take the most important cue first and stop when that cue decides. Stop rules halt search when enough has been learned for the decision at hand. These are not irrational failures of analysis. They are designed frugality.

The distinction matters on large capital projects because the front end constantly invites both kinds of shortcut. A team that generalises from two vivid prior projects is using availability. A team that asks "have we, or has anyone, seen this class of scheme before?" and then looks at the class distribution is using recognition as a useful heuristic. A team that fills every cell of an early estimate because empty cells look unprofessional is practising fake precision. A team that states a parametric range and names what would change it is practising adaptive frugality.

Under time pressure the distinction becomes sharper. Research on decision strategy under acute stress and tight clocks finds that people are pushed toward faster heuristics and can choose poorer strategies when the window closes. That finding is not an argument against heuristics. It is an argument for naming the useful ones in advance, so that when the room heats up the team reaches for a designed rule rather than for whatever story is easiest to tell. A pre-agreed stop rule beats an improvised one. A named class beats a vivid anecdote. A decisive kill criterion beats a scorecard that averages everything into a comfortable middle.

The gate question is therefore not "are we using a shortcut?" Early decisions always do. The gate question is "which kind of shortcut are we using — and do we know it?"

Figure 1: Two kinds of shortcut under scant information. Biased shortcuts build a clean story from what is visible; useful heuristics use few decisive cues, stop rules and honest ranges. The gate question is which kind is in play.

4. Simple Beats Fancy When the Detail Is Fake

At concept stage, the instinct to add detail is strong. Boards ask for more numbers. Advisers supply more models. Packages multiply assumptions until the pack looks thick enough to approve. The uncomfortable finding from judgement research is that this instinct often makes matters worse.

Experts are frequently no better than simple models on prediction tasks, and the amount of experience a person has correlates only weakly with accuracy. Inside estimates — forecasts built from the case at hand, its objectives, its resources and a mental simulation of progress — carry systematic optimism. Deliberation in groups can polarise rather than correct. The practical remedy is not to discard expertise but to stop treating detailed expert narrative as automatically superior to a simple model or a simple outside check.

The same logic applies to early quantification. Parametric methods that relate cost or duration to a few size and type drivers are often more appropriate at concept stage than bottom-up builds that invent quantities the design does not yet support. Premature detail creates false confidence. It also crowds out the strategic questions that scant-information methods are meant to answer: is this the right concept, what would kill it, and how does it sit in a class of similar schemes?

None of this argues that expert judgement is useless. It argues that expertise improves the quality of the inside view without automatically correcting its blind spots, and that elicitation, calibration and the combination of expert opinions are disciplines of their own. Those disciplines belong when probability and risk numbers will actually drive a gate. They are not a reason to inflate early concept packs with invented precision. Simple rules and simple models are legitimate at the front end. They are often the honest form of early analysis.

Figure 2: Detail is not accuracy. Scant cues and heuristic or parametric ranges are the right tools at concept stage; FEED-grade detail belongs after go-ahead. A three-decimal concept estimate is a warning, not diligence.

5. The Outside View as One Heuristic — Not the Whole Method

Among useful early rules, one deserves a named place in the kit without being mistaken for the whole forecasting method.

The outside view treats the project as one of a class of similar efforts. Instead of constructing an elaborate story from the case at hand, it asks what happened to comparable projects, lays out a rough distribution of outcomes and places the current scheme in that distribution. Framed this way, the outside view is a simple statistical check on analogous earlier efforts — a heuristic stance switch that bypasses many of the cognitive and organisational amplifiers of optimism.

On large capital projects that stance is powerful. Uniqueness bias — the belief that this scheme is singular and therefore incomparable — kills base rates before they can speak. The corrective heuristic is short: this project is one of those. Anchor on the class before adjusting, and adjust only with reasons the data can support.

What this blog does not do is rebuild the full reference-class procedure, the institutional mandates that surround it, or the empirical base-rate tables that belong to that method. Those are a method in their own right. The point here is narrower. Before an owner team has the data or the time for a full class study, and especially when the immediate job is concept choice rather than a public cost promise, the outside-view stance still belongs in the room as a fast rule: name a class, refuse uniqueness as a default, and do not let an inside story stand alone beside a go decision.

When a named reference class with usable data does exist, the hand-off is upward: escalate from heuristic stance to the full outside-view method rather than decorating a guess with a generic percentage and calling the work done.

There is a related discipline about optimism's place. Realistic forecasts belong at the moment of commitment. Motivational optimism has a role after a genuine go decision, when the job is delivery rather than selection. Collapsing those moments — using delivery optimism to win a concept gate, or using forecast realism to demoralise a team that has already locked — confuses two jobs. The outside-view heuristic helps keep them apart: use the class to decide whether to commit; do not ask the class to do the work of leadership after commitment.

6. A Small Kit of Early Decision Heuristics

Owner teams need more than a contrast between bias and virtue. They need a small, usable kit for the moments when a concept or early sanction decision cannot wait for FEED. The six rules below are such a kit. They are drawn from the logic of fast-and-frugal heuristics and from the positive leadership heuristics that successful large-project practice keeps returning to. They are not a personality trait and not a complete operating system. They are prompts for the agenda.

Two cautions before the list. First, the kit is for early decisions under scant information. It is not a substitute for detailed planning once a concept has been chosen and the reversible window is being used properly. Second, several neighbouring disciplines — full class forecasting, scenario sets, front-end definition completeness, expert elicitation — remain essential. The kit only claims the strip of work where a few adaptive rules beat both paralysis and fake precision.

1. Class recognition. Ask first whether the team, the owner or the wider industry has seen this class of scheme before. If the answer is yes, pull the class before inventing a unique story. If the answer is no, say so explicitly — unprecedented is a risk statement, not a compliment.

2. Decisive cue first. Name the one or two criteria that should decide before the others: a safety case that cannot be closed, a strategic fit that fails, a fatal downside that no upside story can compensate for. Search lesser cues only after those clear. Lexicographic discipline prevents a crowded scorecard from hiding a kill criterion under average scores.

3. Concept stop rule. Decide in advance what "enough" means for this gate. Enough to choose or kill a concept is not the same as enough to price FEED, freeze a public opening date or publish a single-point return. Stop when the concept decision is supported; refuse to let the pack demand delivery-grade numbers for a concept-grade question.

4. Say no and walk away. Underfunded starts, monument schemes that serve prestige more than the right-hand result, untested technology cast as the delivery path, and disputes that do not serve the project's purpose are candidates for a deliberate no. Focus is a heuristic. Not every attractive scheme deserves to consume the reversible window.

5. Watch the downside. At concept stage, fat-tailed kill risks matter more than upside narratives. A heuristic that asks "what would make this fail badly?" before "what would make this look good?" changes the information the room seeks. Upside can be explored after the downside gate is passed.

6. Why before how much. Put the ultimate result in the right-hand box before the estimate theatre begins. A number without a purpose is not a decision aid; it is a prop. The deeper work of framing and shared purpose belongs elsewhere; the kit only insists on sequence: purpose before precision.

Used together, these rules keep early decisions fast without making them careless. Used knowingly, they also make the next section possible: knowing when to put the kit down.

Figure 3: Early-decision heuristic kit for large capital projects. Six cards for concept and early sanction — class recognition, decisive cue, concept stop rule, say no, watch downside, and why before how much — with a hand-off when another tool owns the job.

7. When a Shortcut Is the Wrong Tool

Heuristics are one exit from scant information, not a universal substitute for every front-end discipline. Owner teams lose quality when they stretch a useful rule into a job it was not built for.

When the problem is irreducible uncertainty about futures — several plausible worlds, not a missing number in one world — the right move is a small set of scenarios that make that uncertainty explicit, not a single heuristic forecast dressed as a base case. When the decision about to be taken is irreversible — a public cost and date narrative, a route or technology freeze, a full-funds authorisation — the right move is to keep learning while learning is still reversible, then compress delivery after lock. When the room is hot with blame, status and clock pressure, the right move is a short judgment protocol that keeps dissent usable tonight, not a casual rule of thumb taken under threat.

When a named reference class with data exists and the gate will publish a commitment number, escalate to the full outside-view method rather than stopping at analogy. When expert probability and risk judgements will actually drive the gate, invest in elicitation and calibration rather than treating a senior opinion as self-validating. When new information arrives that could reorder alternatives, reopen the case instead of defending the last heuristic conclusion. And when the question is whether definition of a chosen concept is complete enough to authorise, that is a front-end loading and quality-at-entry question — a different object from "which adaptive rule helps us choose while numbers are thin."

The map is simple. Ask what job the decision is doing. Then pick the tool that owns that job. Heuristics own frugal concept choice and early stance-setting under scant cues. They do not own scenario sets, irreversible sequencing, hot-room hygiene, full class forecasting, expert combination or the reopening of a case when the information has changed.

Figure 4: Which tool when information is thin. Scant information sits at the centre; exits lead to the heuristic kit, scenario sets, the outside-view method, slow-then-fast sequencing, or elicit-and-reopen — each matched to the job, not used as a universal substitute.

8. Practice on Large Capital Projects

The practices below translate the argument into moves an owner can put on a concept or early-sanction agenda.

Label the decision. Mark every live early item as concept choice, forecast for commitment, or delivery detail. Refuse to treat them as one undifferentiated pack. A concept stop rule cannot work if the pack silently demands a commitment-grade number.

Put class recognition and downside on the agenda before precision. Before any three-decimal figure is debated, ask whether a class has been named and what would kill the scheme. If those questions have no owners, the room is about to optimise a story.

Permit honest early forms. Parametric estimates, ranges and explicit uncertainty statements are diligence at concept stage. Fake precision is a red flag. Reward the team that shows what is unknown.

Separate "enough to choose" from "enough to freeze the public narrative." Choosing a concept can be done with scant information used well. Freezing a public cost and date cannot. Keep the reversible window open until the irreversible story is actually ready.

Escalate when data exist. When a usable reference class is available, move from heuristic stance to full outside-view work. Do not decorate a guess with a generic uplift and stop.

Director and investment-committee prompts. Which shortcut are we using, and is it the biased kind or the useful kind? What would make us stop and switch tools? Are we about to invent precision because the calendar demands a pack, or because the evidence supports it?

On a typical large capital project these moves show up in concrete places. Concept workshops stop treating a preferred scheme as the default and start with class recognition and kill criteria. Sanction packs show ranges and drivers beside any early number, and label which decisions are still reversible. Interface with advisers changes too: the request is not "give us more decimals" but "show us the cues that decided, the class you used, and what would change the call." Contractors and designers joining later inherit a concept that was chosen knowingly, not a fiction that will unravel when FEED finally arrives.

These moves do not slow a project down. They prevent the expensive form of delay: years spent delivering a concept that should have been challenged when challenge was still cheap.

9. Closing: Decide Well Before the Numbers Arrive

Large capital projects will always ask owners to decide before the numbers feel complete. That pressure is structural. Inventing completeness and refusing to decide are both avoidable.

The alternative is disciplined frugality. Treat scant information as the working medium of the front end, not as a defect. Distinguish biased shortcuts from useful heuristics. Prefer simple early models to fake detail. Keep the outside view in the kit as a stance, and escalate to the full method when class data exist. Use a small set of rules — class recognition, decisive cues, a concept stop rule, the courage to say no, attention to downside, and purpose before precision — and hand off when scenarios, sequencing, hot-room judgment, expert elicitation or new information own the next job.

The measure of success is not a thicker early pack. It is a concept that survives contact with later information because it was chosen with eyes open, a forecast that can be defended as a class-aware range rather than as theatre, and a team that knows when to stop searching and when to switch tools. Owner teams that decide this way do not wait for perfect numbers. They decide well before the numbers arrive.

References

• Williams, T. M., Samset, K., & Sunnevåg, K. J. (Eds.). (2009). Making Essential Choices with Scant Information: Front-End Decision Making in Major Projects. Palgrave Macmillan. (Esp. Scheibehenne & von Helversen, Ch. 10 “Useful Heuristics”; Samset, Ch. 2; Williams, Ch. 1; Kirkebøen, Ch. 9; Pugh, Ch. 16; Flyvbjerg, Ch. 8; van der Heijden, Ch. 4; Wright, Bolger & Rowe, Ch. 11; Sunnevåg, Ch. 17.)
• Flyvbjerg, B., & Gardner, D. (2023). How Big Things Get Done. Currency. (Coda: eleven heuristics for better project leadership.)
• Lovallo, D., & Kahneman, D. (2003). Delusions of success: How optimism undermines executives’ decisions. Harvard Business Review, 81(7).
• Merrow, E. W., & Nandurdikar, N. (2018). Leading Complex Projects: A Data-Driven Guide to Leading Multiple Successful Complex Projects. Wiley. (Front-end loading adjacency only.)